J. Fluid Mech. (2014), vol. 751, pp. 570–600.

c Cambridge University Press 2014 570

doi:10.1017/jfm.2014.314

Low-Reynolds-number wakes of elliptical

cylinders: from the circular cylinder to the
normal flat plate

Mark C. Thompson1, †, Alexander Radi1 , Anirudh Rao1 , John Sheridan1 and

Kerry Hourigan1,2
1 Fluids Laboratory for Aeronautical and Industrial Research (FLAIR), Department of Mechanical and
Aerospace Engineering, Monash University, Melbourne, Victoria 3800, Australia
2 Division of Biological Engineering, Monash University, Melbourne, Victoria 3800, Australia

(Received 22 October 2013; revised 20 March 2014; accepted 29 May 2014;

first published online 24 June 2014)

While the wake of a circular cylinder and, to a lesser extent, the normal flat plate
have been studied in considerable detail, the wakes of elliptic cylinders have not
received similar attention. However, the wakes from the first two bodies have
considerably different characteristics, in terms of three-dimensional transition modes,
and near- and far-wake structure. This paper focuses on elliptic cylinders, which span
these two disparate cases. The Strouhal number and drag coefficient variations with
Reynolds number are documented for the two-dimensional shedding regime. There
are considerable differences from the standard circular cylinder curve. The different
three-dimensional transition modes are also examined using Floquet stability analysis
based on computed two-dimensional periodic base flows. As the cylinder aspect ratio
(major to minor axis) is decreased, mode A is no longer unstable for aspect ratios
below 0.25, as the wake deviates further from the standard Bénard–von Kármán state.
For still smaller aspect ratios, another three-dimensional quasi-periodic mode becomes
unstable, leading to a different transition scenario. Interestingly, for the 0.25 aspect
ratio case, mode A restabilises above a Reynolds number of approximately 125,
allowing the wake to return to a two-dimensional state, at least in the near wake.
For the flat plate, three-dimensional simulations show that the shift in the Strouhal
number from the two-dimensional value is gradual with Reynolds number, unlike the
situation for the circular cylinder wake once mode A shedding develops. Dynamic
mode decomposition is used to characterise the spatially evolving character of the
wake as it undergoes transition from the primary Bénard–von Kármán-like near wake
into a two-layered wake, through to a secondary Bénard–von Kármán-like wake
further downstream, which in turn develops an even longer wavelength unsteadiness.
It is also used to examine the differences in the two- and three-dimensional near-wake
state, showing the increasing distortion of the two-dimensional rollers as the Reynolds
number is increased.

Key words: absolute/convective instability, vortex streets, wakes

† Email address for correspondence: [email protected]

 Low-Reynolds-number wakes of elliptical cylinders 571
1. Introduction
The circular cylinder has formed the basis of a large body of fundamental research
on bluff body wakes for over a century starting with the works of Strouhal (1878),
Bénard (1908), Kármán (1911) and numerous studies since (e.g. Roshko 1954; Taneda
1959; Norberg 1987; Monkewitz 1988; Williamson 1988; Eisenlohr & Eckelmann
1989; Strykowski & Sreenivasan 1990; Albarede & Monkewitz 1992; Dusek, Le Gal
& Fraunie 1994; Zhang et al. 1995; Barkley & Henderson 1996; Thompson, Hourigan
& Sheridan 1996; Williamson 1996a,b; Henderson 1997; Thompson, Leweke &
Williamson 2001; Kumar & Mittal 2006; Behara & Mittal 2010; Kumar & Mittal
2012, and many, many, others). This can be attributed to a number of factors
including its symmetry, relevance to industry applications such as oil rig risers,
chimneys, transmission wires, amongst other things and perhaps even the intrinsic
aesthetic beauty of the laminar wake. Of interest to the current paper, the near- and
far-wake states have been explored and documented, as have the three-dimensional
transitions that lead to a turbulent wake as the Reynolds number is increased.
On the other hand, there has been much less research on the wakes of other
slightly less symmetrical bodies such as elliptic cylinders, with this body shape
mainly investigated for very low (Shintani, Umemura & Takano 1983) or high
Reynolds numbers (Modi & Wiland 1970; Modi & Dikshit 1975; Ota, Nishiyama &
Taoka 1987), and even then mostly for non-zero angles of attack.
An investigation by Modi & Dikshit (1975) examined the vortex spacing and
arrangement in the wake of elliptic cylinders for eccentricities in the range
0.44 6  6 0.98 for Re = 68 000. It was found that a lower eccentricity leads to
rapid changes in the longitudinal and lateral vortex spacing in the wake away from
the circular cylinder case. Also focusing on the high-Reynolds-number case, Ota et al.
(1987) examined the wakes of elliptic cylinders in the range 35 000 6 Re 6 125 000 in
terms of the universal Strouhal number introduced by Roshko (1954), based on wake
width and shear-layer velocity, showing that this Strouhal number was insensitive to
angle of attack.
At lower Reynolds numbers, Jackson (1987) used two-dimensional finite-element
simulations and stability analysis to determine the critical Reynolds numbers for
the initial steady-to-unsteady transition. This was done for a range of body shapes
including elliptic cylinders. The critical Reynolds number of a circular cylinder was
determined as 45.4, while for a flat plate it reduces to 27.8. More recent estimates
for a circular cylinder determine the value as 47.3 (Kumar & Mittal 2006) using a
blockage of just 0.5 %, while Saha (2013) estimated the value for a 7 % deep flat
plate as 32.45, although the blockage was much higher at 5 %.
Johnson, Thompson & Hourigan (2004) examined the near-to-far-wake two-
dimensional transitions for elliptic cylinders of various aspect ratios for 75 6 Re 6 175.
For certain Reynolds number/aspect ratio ranges, there was a transition from the
near-wake shedding frequency to a non-commensurate far-wake frequency. The far
wake resembled the standard Bénard–von Kármán (BvK) wake but was formed
from an instability of the time-mean near wake, which had transformed into a
two-layered structure. The secondary shedding could occur either far downstream
after the individual near-wake vortices had diffused and merged so that they were
not identifiable, or prior to this, when they were clearly distinguishable. In either
case, the ratio of the near- and far-wake frequencies was not a simple ratio, and so
the secondary wake was not primarily due to merging of primary vortices. At even
smaller aspect ratios and higher Reynolds number, the secondary wake developed
a characteristic waviness, and even vortex merging in the secondary wake could
572 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan
be observed. These observations are broadly in line with the behaviour of circular
cylinder wakes investigated by Cimbala, Nagib & Roshko (1988), Williamson &
Prasad (1993) and Kumar & Mittal (2012), although for the circular cylinder, the
transition occurs at higher Reynolds numbers and far downstream. In any case, the
introduction of eccentricity to the otherwise circular cylinder leads to a whole range of
wake states not observed in the rotationally symmetric case. More recently, Aleksyuk,
Shkadova & Shkadov (2012) examined a small-aspect-ratio elliptic cylinder, again
showing the development of the secondary wake within 40 diameters for Re = 100.
The normal flat plate can be considered to be an elliptic cylinder with zero aspect
ratio. Owing to its simplicity and rotational symmetry, it has been studied more than
elliptic cylinders, with perhaps the first extensive study by Fage & Johansen (1927),
who determined the vortex spacing in the wake. Even for small Reynolds number
of the order of 100, the near wake undergoes a rapid downstream transition to a
two-layered wake with two lines of like-signed vortices offset on each side of the
centreline. However, less knowledge exists on the nature of this wake in general.
Najjar & Balachandar (1998) used three-dimensional simulations to examine the
wake state at Re = 250. The three-dimensionality was observed to have a spanwise
wavelength of 1.2 plate widths, perhaps reminiscent of mode B for a circular cylinder.
The three-dimensional transitions were examined by Thompson et al. (2006b) using
Floquet stability analysis. They showed that an initial quasi-periodic transition
occurred at Reynolds numbers slightly above Re = 100. This mode has a wavelength
of approximately five diameters. At slightly higher Reynolds numbers, a second
three-dimensional mode became unstable with a much shorter wavelength. That mode
was periodic with the two-dimensional base flow. This second mode appears to have
been observed in the wake of inclined plates up to Re = 350 by Yang et al. (2012).
As discussed above for elliptic cylinders, a secondary vortex street develops
downstream of the primary vortex street (Taneda 1959; Johnson et al. 2004; Saha
2007, 2013), which has been attributed to a hydrodynamic instability of the mean
wake. Apart from viscous diffusion, the geometric arrangement of the vortices in the
vortex street determines whether it is predominately stable or whether it breaks down
into parallel shear flow. A critical value was derived by Durgin & Karlsson (1971) as
h/a = 0.365, where h is the cross-wake spacing between positive and negative wake
vortices and a is the separation between vortices of the same sign. Above this critical
value, the vortices stretch out and align with the downstream direction leading to
a parallel shear flow. In turn, this mean shear flow can become unstable forming a
secondary vortex street. Karasudani & Funakoshi (1994) examined this process further
using a point vortex method to examine the evolution of an idealised street for vortex
spacings of 0.31 and 0.41. For h/a = 0.31, the initial vortices remain isolated, while
for h/a = 0.41 the vortices rotate to align with their major axis in the streamwise
direction and stretch and diffuse until they form a nearly parallel shear flow.
The aim of this study is to examine a variety of aspects of the wakes of elliptic
cylinders, including the circular cylinder and the flat plate as the two extreme cases.
This includes characterising the initial transition to periodic flow, the near- and
far-wake transitions, transition to a three-dimensional wake state, Strouhal number
and drag coefficient variations with Reynolds number, the saturated three-dimensional
mode states for the flat plate, and using dynamic mode decomposition (DMD) to
characterise some of the complex wake states. Most of the analysis is based on
two-dimensional simulations and linear or Floquet stability analysis based on these
two-dimensional simulations; however, a few full three-dimensional simulations are
used to explore special cases. The study proceeds by describing the problem and
 Low-Reynolds-number wakes of elliptical cylinders 573

(a) y (b)

U D, 2a
x

2b 0.75 0.50 0.25 0.10 0

F IGURE 1. (a) Problem definition including relevant dimensions. (b) The elliptic cylinders
considered in this study include the special cases of the circular cylinder with aspect ratio
unity and the normal flat plate which corresponds to a zero-aspect-ratio cylinder.

numerical approaches before examining the results of two- and three-dimensional

simulations. Some results from a parallel experimental program (Radi et al. 2013),
which is not the focus of the current paper, are included for direct comparisons.
Mostly, the analysis and interpretations are framed in terms of known results for the
circular cylinder and, to a lesser extent, the normal flat plate.

2. Numerical approach
2.1. Problem definition
The current paper addresses the flow past two-dimensional elliptic cylinders with the
major axis placed normal to the flow direction, as shown in figure 1. The aspect
ratio is defined by the ratio of the semi-minor (b) to semi-major (a) axis length,
i.e. Ar = b/a, and for convenience for comparisons with the circular cylinder, it is
useful to define 2a = D, the diameter of the circular cylinder. Thus, Ar = 0 corresponds
to a flat plate placed normal to the flow and Ar = 1 corresponds to a circular cylinder.
The Reynolds number is based on the major axis length, i.e. Re = U(2a)/ν, where
U is the oncoming uniform flow speed and ν is the kinematic viscosity. This paper
concentrates on the two- and three-dimensional wake transitions as the aspect ratio
changes. There is a parallel experimental program that used flow visualisation and
particle image velocimetry to examine these flows, the results of which will appear
elsewhere (Radi et al. 2013); however, limited experimental comparisons will be given
with the numerical results that form the basis of this paper.

2.2. Numerical method

The flow is solved numerically using a two-dimensional spectral-element method,
which has been validated in many previous related studies (e.g. Thompson et al. 1996,
2006a; Thompson, Leweke & Hourigan 2007; Griffith et al. 2008, 2010; Stewart et al.
2010; Leontini, Lo Jacono & Thompson 2013; Rao et al. 2013a), and so will only be
briefly described here. The computational domain is broken up into a macro-element
mesh, with elements that can have curved sides. Each macro-element is internally
subdivided into a nx × ny nodes, spaced according the Gauss–Legendre–Lobatto
quadrature points and, within each element, the velocity components and pressure
are represented by the tensor product of Lagrangrian interpolating polynomials.
Importantly, the method has been shown to possess spectral spatial convergence
574 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan
as the polynomial order is increased. In the current implementation, a three-step
fractional-step method (e.g. Chorin 1968; Karniadakis & Sherwin 2005) is used for
the temporal discretisation, which is formally second-order accurate.
A few simulations were performed using a three-dimensional spectral/spectral-
element code to investigate the three-dimensional saturated flow states for the normal
flat plate. This parallel code is based on the two-dimensional spectral-element code,
but uses a Galerkin Fourier expansion in the spanwise direction (e.g. see Karniadakis
& Triantafyllou 1992; Thompson et al. 1996).
The boundary conditions were specified as uniform flow (U, 0) at the inflow and
side boundaries, which were positioned 50 major axis lengths from the cylinder, and
constant pressure and zero normal velocity gradient at the outflow boundary. The
cylinder surface was a no-slip boundary. At all boundaries except for outflow, a
first-order normal pressure gradient condition was used as specified in Karniadakis,
Israeli & Orszag (1991). This enforces mass conservation at these boundaries and
leads to overall second-order accurate velocity fields.

2.3. Stability analysis

Linear and Floquet stability analysis is used to determine the critical transitions as the
Reynolds number is increased. This employs the Arnoldi method based on a Krylov
subspace of evolved fields, as described by Mamun & Tuckerman (1995) and Barkley
& Henderson (1996), and has been utilised for a number of similar problems where
more details can be found (Ryan, Thompson & Hourigan 2005; Leontini, Thompson
& Hourigan 2007; Chaurasia & Thompson 2012; Rao et al. 2013b).

2.4. DMD
DMD is also used as part of the analysis. This employs a time sequence of fields,
in this case velocity fields, to construct a spectral decomposition. This process gives
a set of approximations to the Koopman modes, called Ritz vectors, with associated
eigenvalues, called Ritz values. The Ritz vectors are a set of complex spatial modes
which vary sinusoidally in time at frequencies specified by the argument of the
complex eigenvalues. The modes may also grow or decay exponentially in time,
according to the magnitude of the associated eigenvalue. The analysis is particularly
useful to study behaviour controlled by multiple time scales or frequencies, in order
to decouple the spatial behaviour of the different frequency components. Another
important point, unlike a standard discrete Fourier transform taken at a point, is that
it can extract accurate oscillation frequencies even with a relatively short sequence
of fields on which to base the analysis. The approach used here follows closely that
fully detailed in Rowley et al. (2009) and Schmid (2011), so only a very brief outline
is given here.
If x 0 , x 1 , . . . , x m represent a sequence of column vectors (e.g. the velocity fields)
corresponding to times t0 , t0 + 1t, . . . , t0 + (m − 1)1t, then it is clear that subsequent
fields can be related to previous fields through the matrix equation

0 0 · · · 0 α1
 
1 0 · · · 0 α 
 0 1 · · · 0 α2 
x 1 x 2 · · · x m = x 0 x 1 · · · x m−1  (2.1)
  
3  (≡ K C).
. . . 
 .. .. .. 
0 0 · · · 1 αm
 Low-Reynolds-number wakes of elliptical cylinders 575

(a) (b)

F IGURE 2. (a) Mesh used for the Ar = 0.25 cylinder. The other meshes are similarly
configured. The inflow length is 50D and the outflow length 280D. (b) Shows a zoomed
in view of the mesh in the neighbourhood of the cylinder. Only macro-elements are shown,
which are further subdivided internally.

The right-hand side matrix is called the companion matrix, C, and, since all of
the columns vectors x 0 , . . . , x m are known, the unknown terms α1 , α2 , . . . can be
determined by solving an over-determined matrix equation using singular-value
decomposition (SVD) or QR decomposition. Then the m eigenvalues (λk ) and
eigenvectors (zk ) of C can be used for a modal reconstruction of the original data set
in the form
1 λ1 λ21 · · · λ1m−1
 
2 m−1 
1 λ2 λ2 · · · λ2 

x 0 x 1 · · · x m−1 = v 1 v 2 · · · v m  .  (≡ V T). (2.2)
  
.. . .
 .. . . 
2 m−1
1 λm λm · · · λm
Here, the column vectors v k , called Ritz vectors, are just the column vectors of V =
K Z, with Z the matrix formed from the column vectors z0 , . . . , zm−1 . Note that both
the Ritz vectors and values consist of complex conjugate pairs. Equation (2.2) just
states that the field at any discrete time tk can be expressed as a modal expansion
m
λki v i .
X
xk = (2.3)
i=1

If the sequence of fields is strictly periodic, so that x m = x 0 , then the expansion given
in (2.3) is equivalent to a Fourier decomposition.

2.5. Mesh and domain considerations

A typical mesh is shown in figure 2 for an Ar = 0.25 aspect ratio cylinder. The
computational domain extends between −50 6 x/D 6 280 in the streamwise direction
and −50 6 y/d 6 50 in the cross-stream direction. This means blockage is restricted
to 1 %. The long outflow length allows various two-dimensional wake transitions
to occur for the smaller-aspect-ratio bodies. Note that resolution is maintained
towards the outflow boundary to resolve the wake fully over the entire domain.
For the production runs, each internal element is subdivided into 6 × 6 node points;
increasing the internal resolution to 8 × 8 resulted in a change in the near-wake
Strouhal numbers and the Floquet multipliers for the Re = 200 cases of less than
0.01 and 1 %, respectively. Six different aspect ratios were considered: Ar = 0.0, 0.10,
0.25, 0.50, 0.75 and 1.0. The Reynolds number range was restricted to Re 6 200.
576 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

Ar Rec Stc
(Flat plate) 0.00 31.6 0.0998
0.10 33.5 0.1030
0.25 35.6 0.1074
0.50 38.8 0.1120
0.75 42.6 0.1144
(Circular cylinder) 1.00 47.2 0.1163
TABLE 1. Critical Reynolds numbers and Strouhal numbers for the initial
steady-to-unsteady-wake transition for elliptic cylinders of different aspect ratios.

3. Results
3.1. Transition to unsteady flow
As for the circular cylinder, elliptic cylinders also undergo an initial Hopf bifurcation
from a steady to periodic wake flow. This takes place at Reynolds numbers lower
than for the circular cylinder, as was shown by Jackson (1987). Table 1 shows the
transition Reynolds numbers for the different aspect ratio elliptic cylinders considered.
These were from simulations at subcritical Reynolds numbers starting from saved
solutions at slightly supercritical Reynolds numbers. The flow was evolved until the
oscillation amplitude was decaying purely exponentially, allowing the decay rates to
be accurately determined. The critical values could then be determined very accurately
by extrapolation to zero growth rate.
For a flat plate, Saha (2007) finds a value of Rec = 32.5, while the early numerical
study of Jackson (1987), with a 10 % blockage, found values of Rec = 27.8 and
Stc = 0.124. The Hopf bifurcation for a circular cylinder has been examined by
a number of authors: Rec = 45.4 (Jackson 1987), 46.1 (Dusek et al. 1994), 46.7
(Le Gal, Nadim & Thompson 2001), 47.3 (Kumar & Mittal 2006), 49 (Williamson
1996b, experimental). For the numerical studies, the differences are mainly due to the
different blockage ratios of the grid systems used. The result from Kumar & Mittal
(2006) employed the lowest blockage ratio (0.5 %), and hence is the most accurate.
(The blockage ratio for the current study is 1 %.) In addition, Kumar & Mittal (2006)
gives the critical Strouhal number for the circular cylinder transition to be 0.1163,
which was the value found in the current study to four significant figures. Hence, the
results given in table 1 are in excellent agreement with previous studies.

3.2. Two-dimensional shedding

Aspects of the two-dimensional wake transitions have been considered by a number of
authors, both experimentally and numerically: Cimbala (1984), Cimbala et al. (1988),
Williamson & Prasad (1993), Johnson et al. (2004) and Saha (2013), although mainly
for a circular cylinder. Johnson et al. (2004) examined the transitions using Fourier
analysis of time sequences at points along the centreline. That work is extended in
this paper. Figures 3 and 4 show the asymptotic wake behaviour in terms of wake
vorticity patterns as the aspect ratio and the Reynolds number are varied. For the
circular cylinder at Re = 100, the near wake is the standard aligned BvK street,
which further downstream develops into a two-layered wake, with clockwise vortices
forming a line offset from the anticlockwise vortices below. Yet further downstream,
the clockwise and anticlockwise vortices diffuse and cross-annihilate, leaving a
 Low-Reynolds-number wakes of elliptical cylinders 577

(a)

(b)

F IGURE 3. (Colour online) Asymptotic wake states for different aspect ratios as the
Reynolds number is varied using the vorticity field to highlight the structure: (a) Re = 50;
(b) Re = 100. The aspect ratios are displayed on the plots. The first 250 diameters of the
wake are shown. The flow is from left to right. Blue (darker grey) and red (lighter grey)
denote positive and negative vorticity, respectively.

mean wake structure without identifiable individual vortices. Increasing the Reynolds
number further to Re = 200 means that this evolved mean wake persists much further
downstream due to the reduced viscous damping. There is no sign of a secondary
wake, despite its appearance in other numerical studies (Kumar & Mittal 2012); the
higher accuracy of the spectral-element method presumably leads to less numerical
noise to trigger the convective instability. The two-layer wake that develops from the
near wake has a reduced streamwise wavelength. For the next smaller aspect ratio
578 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

(a)

(b)

F IGURE 4. (Colour online) As for figure 3 with (a) Re = 150 and (b) Re = 200.

cylinder (Ar = 0.75), the same sequence of transitions occur but for lower Reynolds
numbers, or equivalently closer to the cylinder for the same Reynolds number. This
is presumably due to the increased strength of the shed vortices, which will be
quantified later. For smaller aspect ratios, the transition process is further accelerated.
For instance, for Ar = 0.5 at Re = 200, the transition from the BvK street to the
 Low-Reynolds-number wakes of elliptical cylinders 579

22

20

18

16

14

12

10
50 75 100 125 150 175 200
Re

F IGURE 5. Integrated vorticity (i.e. circulation) passing though the top shear layer into
the wake from vorticity generated at the front of the cylinder as a function of Reynolds
number. The aspect ratios are marked.

two-layered street occurs within the first two diameters of the cylinder, and the
stationary wake profile does not form at all. Instead, the identifiable vortices in the
two-layer wake interact at approximately 20D downstream to form a much longer
wavelength secondary vortex street, which itself oscillates spatially with an even
longer wavelength. As the aspect ratio is reduced still further, the far wake becomes
increasingly complex and chaotic. For Ar = 0.1 and Re = 200, vortex pairs form as
part of the far wake and self-propel away from the centreline. The normal flat plate
wakes, not shown, are similar to the Ar = 0.1 wakes.
Part of the reason for the different behaviour can be attributed to the increasing
vorticity flux feeding into the wake for the smaller-aspect-ratio cases, as the flow
has a higher speed as it passes over the top of the cylinder. This is quantified in
figure 5, which gives the circulation entering the wake each shedding cycle as the
flow separates near the top of the cylinder, and feeds the vorticity generated at the
front of the cylinder into the wake. This circulation is given by
Z Z ∞ 
Γ= uω dy dt. (3.1)
one period 1/2

(Note that non-dimensional variables are used.) Clearly, the circulation convecting into
the wake is considerably higher for the flat plate compared with that for a circular
cylinder. At Re = 200, the difference is almost a factor of two. The correspondingly
larger circulation per wake vortex has a significant effect on both the near- and far-
wake dynamics as is demonstrated in figures 3 and 4.

3.3. Temporal evolution of the wake for small aspect ratio

Figure 6 shows the development of the wake through the transient stage for the Ar =
0.25 cylinder at Re = 150. The initial evolution is similar to that for a circular cylinder,
i.e. the development of a long recirculation zone at the rear of the cylinder which
becomes unstable and starts to shed vortices. The initial wake then resembles the
580 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

F IGURE 6. (Colour online) Transient development of the wake for Ar = 0.25 at Re = 150.
The images correspond to non-dimensional times: τ ≡ tU/D = 100, 150, 175, 200 and 375.
The images are truncated at x/D = 50.

standard aligned BvK vortex street. However, after this stage the development deviates.
A few diameters downstream of the cylinder, the vortices begin to bunch up, showing
the beginnings of a two-layered wake. This is shown in the third image of figure 6,
corresponding to a non-dimensional time of 150 units or approximately 50 units after
the attached recirculation bubble begins to develop asymmetries. After another 25 time
units, the two-layered wake structure is clearly apparent with the secondary wake
further downstream formed from merging of those individual two-layer vortices. After
a much longer time, the downstream extent of the two-layered wake has increased and
the secondary wake structure has become more regular.

3.4. Transition to a parallel shear flow

The work of Durgin & Karlsson (1971), Tsuboi & Oshima (1985) and Karasudani &
Funakoshi (1994) investigated the stability of the geometrical arrangement of vortices
in a wake showing that when the spacing ratio h/a was greater than 0.365, the
vortices would rotate to align with the streamwise direction, and stretch out and
diffuse resulting in a near-parallel shear flow. If it forms, this near-stationary wake
would be convectively unstable and could form a new secondary vortex street with
a longer wavelength and lower Strouhal number. Figure 7 plots the variation of the
spacing ratio with downstream distance for a number of different aspect ratio elliptic
cylinders and Reynolds numbers. Each of these wakes exceed the theoretical transition
criterion at some point downstream. The smaller-aspect-ratio ellipses exceed it earlier
and by a larger margin. This is consistent with the observed wakes in figures 3
and 4. The images on the right of figure 7 show that the first 50 diameters of the
wakes for the cases examined, with the position where the spacing ratio is equal to
0.365 marked with dashed lines. Clearly the stability criterion appears broadly correct,
 Low-Reynolds-number wakes of elliptical cylinders 581

(a) 1.0 (b)

0.8

0.6

0.4

0.2

0 10 20 30 40 50 60

F IGURE 7. (Colour online) (a) Downstream variation of the vortex spacing ratio (h/a) for
different elliptic cylinders and Reynolds numbers. The dashed line corresponds to a vortex
spacing ratio of h/a = 0.365. (b) The corresponding wakes. The vertical dashed lines show
the approximate positions where h/a = 0.365.

(a) 0.5 (b) 0

Centreline velocity deficit

0.4
−5

0.3
50
−10
100
0.2
150
−15 200
0.1
250
−20
0 50 100 150 200 0 0.05 0.10 0.15 0.20 0.25 0.30
St

F IGURE 8. (a) Comparison of wake deficit (1 − u/U∞ ) for the circular cylinder at Re =
◦
150 with previous experimental results: , Williamson & Prasad (1993); 4, Cimbala et al.
(1988). (b) Power spectra at various downstream positions showing that the BvK spectral
peak is still dominant, although small, even at x/D = 250.

although it may be some distance downstream before the parallel shear flow develops.
For the Ar = 0.25, Re = 150 case, the secondary wake develops well before this
happens, with the secondary vortices forming from direct merging of a non-integer
number of identifiable primary vortices.

3.5. Variation of mean centreline wake velocity

For a circular cylinder, the variation of the centreline velocity with downstream
distance has been determined experimentally by Cimbala (1984), Cimbala et al. (1988)
and Williamson & Prasad (1993). In the two-dimensional shedding regime at Re = 150,
the velocity deficit measured by these authors is shown in figure 8(a), together with
predictions from the current simulations. In general, the numerical predictions of this
deficit are approximately 10 % higher than the results of Williamson & Prasad (1993)
582 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

(a)

(b)

F IGURE 9. (Colour online) (a) The development of the vortex street over the first
50 diameters downstream. (b) Corresponding coloured (greyscale in print) contours of
horizontal velocity component. Here, Re = 150, Ar = 1.0.

and approximately 5 % above those of Cimbala (1984), but with the overall shape
of the distribution predicted very well. Notably, the results from Williamson &
Prasad (1993) do not show the small bump in the distribution approximately 10D
downstream. This small bump seems to corresponds to the transition of a spaced-out
BvK vortex street to a more bunched-up street with more diffused vortices, which
occurs at approximately x/D = 10, as shown in figure 9(a). Matching contours of
period-averaged horizontal velocity component are shown in figure 9(b), showing
the effect is relatively subtle on this view. Both images show the wake to 50D
downstream. The transition to a two-layered wake structure further downstream at
approximately x/D = 45 corresponds to the main peak in the velocity deficit shown
in figure 8(a).
From figure 3, it is clear that even at Re = 200 (slightly above the Reynolds number
of three-dimensional transition), the secondary wake sets in a long way downstream
(x/D > 200) for the circular cylinder. In fact, for the current simulations, there is no
sign of this development in the vorticity plots. However, power spectra at different
distances downstream show that the secondary shedding frequency begins to become
detectable at x/D ' 200, and even then the far-wake spectral peak at St = 0.0723 is
still more than two orders of magnitude below the BvK peak at x/D = 250 (figure 8b).
As indicated above, Kumar & Mittal (2012) examined this case through numerical
simulations recently. They found the far-wake spectral peak began to dominate the
BvK peak for x/D & 150. They also found that reducing the time step also reduced
the numerical error, which in turn reduced the far-wake spectral peak and increased
the downstream position for effective onset. On the other hand, upstream noise in
experiments generally causes onset even earlier. For instance, Vorobieff, Georgiev &
Ingber (2002) using a flowing soap film, found the onset of the secondary wake to
occur at x/D ' 80. The centreline velocity profiles for other aspect ratios are shown
in figure 10. These show substantially different behaviour as the Reynolds number is
varied, in line with the image sets shown in figures 3 and 4. A decrease in aspect
ratio is clearly associated with an increase in the wake deficit on the centreline. For
example, for Ar = 0.5 at Re = 150, there is a region where the centreline velocity
is less than 10 % of the free stream for 5 . x/D . 20, corresponding to where the
wake has a two-layered structure. This is apart from the near-wake mean recirculation
zone at the start of the wake. As the body is deformed towards a flat plate, this
effect is amplified, as recently shown by Saha (2013). For an elliptical cylinder with
Ar = 0.25, the centreline mean wake velocity is almost zero between 3 . x/D . 20 for
Re = 150. This case is interesting, as it will be shown later, in that the wake is still
two-dimensional at this and higher Reynolds numbers for this aspect ratio.
 Low-Reynolds-number wakes of elliptical cylinders 583

(a) 1.0 (b) 1.0

150
0.8 200 0.8
Centreline velocity

0.6 50 0.6
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4 −0.4
0 10 20 30 40 50 0 10 20 30 40 50

(c) 1.0 (d) 1.0

0.8 0.8
Centreline velocity

0.6 0.6
0.4 0.4
0.2 0.2
0 0
−0.2 −0.2
−0.4 −0.4
0 10 20 30 40 50 0 10 20 30 40 50

F IGURE 10. Mean velocity on the centreline as the Reynolds number is varied (Re = 50,
100, 150, 200): (a) Ar = 0.0; (b) Ar = 0.25; (c) Ar = 0.50; (d) Ar = 1.0. Solid line, Re = 50;
long dash, Re = 100; short dash, Re = 150; medium dash, Re = 200.

3.6. Strouhal–Reynolds number variation

The Strouhal–Reynolds number variation for a circular cylinder has been documented
by a number of authors, e.g. Roshko (1954), Barkley & Henderson (1996) and
Williamson (1996a), with Williamson (1996b) citing a more complete list. Figure 11
shows the St–Re variations obtained from two-dimensional simulations for the set
of elliptical cylinders studied. The curve for Ar = 0.75 is only slightly shifted from
the circular cylinder curve. The lower-aspect-ratio curves show distinctly different
behaviour, reaching a maximum Strouhal number at increasingly smaller Reynolds
numbers as the aspect ratio is reduced. For low Reynolds numbers (Re 6 60), the
curves appear to approach an envelope curve as the aspect ratio is reduced.
Figure 12 shows a comparison of the numerically predicted Strouhal/Reynolds
number variations against those obtained from experiments (Radi et al. 2013). The
four aspect ratio pairs shown do not match perfectly, nevertheless they are relatively
close. (The cylinders used in the experiments were designed to have the same aspect
ratios; however, the measured values of the manufactured cylinders were slightly
different.) Also note that the Are = 0.07 aspect ratio cylinder is actually a sharp-edged
rectangular cylinder with an aspect ratio of 7 %.
There are a number of things to note about these comparisons. For the circular
cylinder, the numerical and experimental curves match to better than 1 % up to
Re ' 177, at which point the experimental flow becomes three-dimensional as the
flow undergoes mode A transition. During this hysteretic transition, the Strouhal
number drops considerably, consistent with other experiments (e.g. Williamson 1996a).
584 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

0.22
0.75
0.20 1.00

0.18 0.50
0.25
0.16
St *0.10
0.14 *

0.12

0.10

0.08
25 50 75 100 125 150 175 200
Re

F IGURE 11. Strouhal number variation with Reynolds number from two-dimensional
simulations. Each curve corresponds to a different aspect ratio as marked. The asterisks
correspond to approximate Strouhal numbers for the flat plate (Ar = 0) obtained from
direct three-dimensional simulations.

0.20

(0.75, 0.72) (1.00, 1.00)

0.18

(0.25, 0.26)
St 0.16

0.14

0.12

100 120 140 160 180 200

Re

F IGURE 12. Comparison of numerical (solid) and experimental (dashed)

Strouhal/Reynolds number variations. Here, (Arn , Are ) indicates (pairs of similar)
numerical and experimental aspect ratios.

On decreasing Reynolds number, the transition from three-dimensional flow back to

two-dimensional flow occurs at Re ' 169. For the next-lowest aspect ratio pair,
(Arn , Are ) = (0.75, 0.72), the three-dimensional transition occurs at a considerably
lower Reynolds number, and the hysteric range is much smaller: 130 6 Re 6 132. As
for the circular cylinder, the experimental and numerical Strouhal number curves are
still very close in the two-dimensional shedding regime. For the third aspect ratio pair,
(Arn , Are ) = (0.25, 0.26), perhaps surprisingly, the curves match better at the higher
Reynolds number range 120 6 Re 6 200, and deviate at lower Reynolds numbers. The
final aspect ratio pair, (Arn , Are ) = (0.10, 0.07), shows good agreement between the
 Low-Reynolds-number wakes of elliptical cylinders 585

3.0

0
2.5

0.25

CD 2.0
0.50

1.5

1.0
50 75 100 125 150 175 200
Re

F IGURE 13. Drag coefficient variation with Reynolds number from two-dimensional
simulations. The aspect ratio is marked. The dashed line corresponds to the numerical
results from Henderson (1995) for the circular cylinder. The discrepancy is probably due
to slightly higher blockage of the Henderson (1995) results, although the blockage was not
stated in that paper. The asterisks correspond to a 7 % width flat plate with 5 % blockage
(Saha 2013). The current blockage ratio is 1 %.

predictions and the measurements for 100 6 Re 6 170, beyond which the two curves
begin to move apart.

3.7. Drag–Reynolds number variation

The variation of drag coefficient with Reynolds number is given in figure 13 for
aspect ratios: 0, 0.25, 0.5 and 1. The benchmark results of Henderson (1995) are
overlaid for the circular cylinder, showing a match up of better than 2 % over most
of the range. The discrepancy is most likely due to a slightly higher blockage ratio
for the previous results, although the blockage is not given in Henderson (1995). In
addition, some results from Saha (2013) are provided for the flat plate. For those, the
blockage was 5 %, and the flat plate had 7 % thickness, so it is not surprising there is
a greater difference in the predictions for this geometry. The lower-aspect-ratio cases
show higher drag and also more complex variation with Reynolds number.

3.8. Three-dimensional transition

Similar to the study of Barkley & Henderson (1996), Floquet analysis was used to
determine the initial three-dimensional transitions. The critical Reynolds numbers for
(the equivalent of) the mode A transition is given in table 2.
The critical Reynolds number decreases significantly with aspect ratio, dropping
from Rec,A = 190.3 to 88.2 as the aspect ratio is reduced from Ar = 1 to 0.25. The
critical Reynolds numbers from experiments are somewhat lower than the numerical
results: Rec,A = 177 against 190.3 for Ar = 1.0, and 132 against 148.6 for Ar = 0.72
(0.75). Barkley & Henderson (1996) found Rec,A = 188.5 ± 1 also using Floquet
analysis, which is close to the value found here, again with the difference likely
to be due to slightly different blockage ratios. The lower critical Reynolds numbers
from the experiments are consistent with other results for a circular cylinder, where
586 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

Ar Numerical Experimental
Rec,A λc,A Rec,A λc,A
(Flat plate) 0.00 — —
0.10 — —
0.25 88.2 5.34
0.50 112.2 4.38
0.75 148.6 3.92 132 ∼4
(Circle) 1.00 190.3 3.98 177 ∼4
TABLE 2. Critical Reynolds numbers and corresponding wavelengths for mode A
transition.

(a) (b)

(c) (d )

F IGURE 14. (Colour online) Comparison of the mode A streamwise vorticity structure as
the ellipse eccentricity is varied showing that the near-wake perturbation field structure is
essentially the same: (a) Ar = 1.0, Re = 190, λ/D = 4; (d) Ar = 0.75, Re = 150, λ/D = 4.0;
(a) Ar = 0.50; Re = 110, λ/D = 4.4; and (d) Ar = 0.25; Re = 90, λ/D = 5.2.

background perturbations and end effects can lead to similar early transition (e.g.
Williamson 1988, 1989), although Miller & Williamson (1994) found the transition
could be delayed beyond Rec,A = 200 using contaminating end conditions.
The preferred wavelength at transition λc,A only varies weakly over this aspect ratio
range, increasing, although not quite monotonically, from 3.98 to 5.34. Interestingly,
mode A remains subcritical for the two smallest aspect ratios. For Ar = 0.1, the
Floquet multiplier attains a value of µ ' 1 at (Re, λ) = (85, 6.0), indicating neutral
stability, but decreases at higher Reynolds numbers. Hence, if it is unstable at all, it
is only marginally unstable and, if so, for only a narrow range of Reynolds numbers
before it restabilises.
Evidence that the initial transition is effectively equivalent to the same instability
mode as mode A, previously identified for a circular cylinder wake, is shown in
figure 14. This shows the streamwise perturbation vorticity field near onset for aspect
ratios Ar = 1.0, 0.75, 0.50 and 0.25. Clearly, in the near-wake the mode structure
is retained, as the aspect ratio (and Reynolds number) varies considerably. For the
smaller-aspect-ratio cases, the wake, shown by the thin black lines, undergoes a rapid
transition to a two-layered wake structure, as can be seen towards the downstream
end of the images. For still smaller aspect ratios (Ar = 0.1 or 0), the wake does
not experience the mode A transition, possibly because the two-layered wake moves
even further upstream, which presumably modifies the near wake sufficiently so that
 Low-Reynolds-number wakes of elliptical cylinders 587

(a) (b)

F IGURE 15. (Colour online) (a) Snapshots of the spanwise vorticity field of the flat plate
wake at Re = 120 at one (near-wake) period apart. The downstream transition to secondary
shedding prevents the wake being exactly periodic. Nevertheless, over the near-wake region
marked by the box, the flow is very close to periodic. The Floquet analysis was performed
by evolving the flow in addition to the perturbation fields and restricting the calculation
of Floquet multipliers to the boxed region only. (b) Streamwise perturbation vorticity for
the initial transition mode for the flow past a flat plate at Re = 120, close to the transition
Reynolds number. Even the near-wake structure is very far from the standard BvK
wake of a circular cylinder and there is little resemblance to the mode A wake structure.
The mode appears to be reasonably close to subharmonic, with a Floquet multiplier
of 1.08e0.78πi .

mode A can no longer reach positive growth. These results were obtained using an
outflow domain length of 20D, which was short enough to ensure that the wake
was purely periodic, allowing standard Floquet stability analysis to be performed.
Increasing the outflow length, especially for the lower-aspect-ratio cases, allows the
wake to undergo secondary transition further downstream with a different frequency
than the near-wake shedding frequency. This is discussed in detail later in the paper.
For the zero-thickness flat plate, the initial transition is to a complex Floquet mode,
meaning that the frequency of the mode is different from the wake frequency. This
mode was previously identified in Thompson et al. (2006b). It becomes unstable at
Rec ' 116 for a spanwise wavelength of λc ' 5.0D. The critical Reynolds number
given here is slightly higher than that suggested in Thompson et al. (2006b) of
Rec = 105–110. In that study, a short outflow length was used to enforce exact wake
periodicity, by preventing the transition to secondary shedding further downstream.
That seems to have had the effect of reducing the critical Reynolds number slightly.
For the case here, a longer domain is used, with an outflow length of 50D. This causes
the wake velocity signal to no longer be strictly periodic even in the near wake. Given
this, Floquet stability analysis cannot be strictly applied; however, it can be applied in
an approximate sense by determining the growth multipliers over the average period.
Taking the Re = 120 case as an example, the measured period at x = 1.5D downstream
on the centreline varies by approximately 0.3 % between approximately 7.17 and 7.19
over different periods. This is due to the downstream transition to a secondary
shedding mode having an upstream effect on the near wake. As well as using the
average near-wake period, it is also necessary to restrict the analysis by only taking
account of the fields in the near wake, and not including the downstream wake
where secondary transition has already occurred, to compute the (pseudo-)Floquet
multipliers (see figure 15a). This procedure resulted in the computation of Floquet
multipliers that typically varied by less than 0.1 % from cycle to cycle. The streamwise
perturbation vorticity field for this mode is shown in figure 15(b), demonstrating that
the perturbation field structure is quite different to mode A.
588 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

F IGURE 16. (Colour online) Streamwise perturbation vorticity field corresponding to

mode A for Re = 120 and λ = 4.6D.

(a) (b) (c)

F IGURE 17. (Colour online) Streamwise perturbation vorticity field corresponding to

mode AA for Re = 140 for the Ar = 0.00 cylinder (a) and Re = 190 for Ar = 0.25 (b). The
right-hand image (c) shows the corresponding spanwise perturbation vorticity field for the
Ar = 0.25 cylinder at Re = 190.

3.9. Three-dimensionality past transition

Decreasing the aspect ratio results in progressively more rapid downstream transition
to the two-layered vortex street. For the Ar = 0.25 aspect ratio cylinder, mode A
initially becomes unstable at Re = 88; however, the mode is no longer unstable beyond
Re ' 120–125. Figure 16 shows the streamwise perturbation vorticity field at Re = 120
for λ = 4.6D. At this point it is still unstable with a Floquet multiplier of 1.04, but this
drops to 0.91 by Re = 130. Comparing this image with those in figure 14 indicates that
the wake undergoes a rapid transition to a two-layered wake close to the back of the
cylinder. As this transition migrates upstream with increasing Reynolds number, the
development of the mode A instability is suppressed. This is similar to the situation
with the flat plate shown in figure 15(a), where mode A does not became unstable
at all. Hence, at least in the near wake, there is a transition from a three-dimensional
wake back to a two-dimensional wake. This ties in very well with the experimental
result from figure 12, which shows that the Strouhal number curve beyond Re = 120
obtained from experiments (Radi et al. 2013) tracks the two-dimensional numerical
prediction closely. Experimental flow visualisations also show that three-dimensionality
in the near wake is suppressed at these Reynolds numbers, whilst the near wake is
clearly three-dimensional at lower Reynolds numbers (Radi et al. 2013).
For Ar = 0.25, the wake remains two-dimensional beyond Re = 125 until another
mode becomes unstable at Re = 185 (see table 3). The streamwise vorticity
perturbation contours for that mode are shown in figure 17(b) and the spanwise
contours in figure 17(c). The spatiotemporal symmetry is the same as mode A but
the near wake and mode structure within the two-layered wake vortices are different.
In particular, the spanwise vorticity contours show the characteristics of elliptic
instability in these vortices, with a radial mode number of two, i.e. two radial nodes
 Low-Reynolds-number wakes of elliptical cylinders 589

Ar First mode Second mode

Rec,QP λc,QP Rec,AA λc,AA
(Flat plate) 0.00 116 5.0D 141 2.1D
0.25 — — 185 1.3D
TABLE 3. Critical Reynolds numbers and corresponding wavelengths for the initial
quasi-periodic mode QP transition, and also mode AA transition, for small-aspect-ratio
cylinders.

(a) (b)

F IGURE 18. Wake vorticity structure for a flat plate, highlighted using the criterion of
Jeong & Hussain (1995) with λ2 = −0.2 from full three-dimensional simulations. Span
shown is 16D, using 128 Fourier planes: (a) Re = 140; (b) Re = 180.

in the perturbation field. (For mode A for Ar = 0.25 shown in figure 16, the radial
mode number in the two-layer wake vortices is one.) This is consistent with the much
shorter preferred wavelength of this new mode of λ = 1.3D at onset. Interestingly,
the equivalent mode is observed for the flat plate wake. This mode becomes unstable
at Re = 141 (see table 3), and has a spanwise wavelength of 2.1D. Note that this is
a higher critical Reynolds number than reported previously (Thompson et al. 2006b),
again because of the stronger influence of a shorter outflow length than expected, as
discussed previously. The preferred wavelength at onset is considerably larger than
that for the Ar = 0.25 cylinder of 1.3D. However, this appears consistent with the
larger size of the wake vortices for the flat plate, as can be seen in figure 17(a,b).
For the flat plate, the quasi-periodic mode remains unstable for this Reynolds number,
and remains so for higher Reynolds numbers. Hence, the flat plate wake does not
return from a three-dimensional state to a two-dimensional state as the Reynolds
number is increased.
Figure 18 shows isosurface visualisations from full three-dimensional simulations
of the flat plate wake at Re = 140 (a) and Re = 180 (b). The isosurfaces correspond
to λ2 = −0.2 using the method of Jeong & Hussain (1995). It shows that increasing
distortion of the spanwise vortical structures with increasing Reynolds number, and
the increasing dominance of streamwise vortices. The transition to a saturated three-
dimensional wake increases the Strouhal number considerably: for Re = 140, St2D '
0.138 and St3D ' 0.142; and for Re = 180, St2D ' 0.136 and St3D ' 0.151. Again, this
is consistent with the experimentally observed Strouhal number increase for a finite-
thickness flat plate shown in figure 12.
590 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

3
f0

2 f3 f2−f3

f2 f1
2f2
1
log10 A
2f1
3f2
4f2
3f1
0 4f1
5f1

−1

−2
0 0.2 0.4 0.6 0.8 1.0
St

F IGURE 19. Frequency spectra from the DMD analysis for Ar = 0.25 and Re = 150. See
the text for details.

3.10. DMD of wakes

Figures 3 and 4 show that these cylinder wakes can undergo various transitions as they
advect downstream. For example, the wake for the Ar = 0.25 cylinder at Re = 150
shows the classical BvK wake very close to its base which rapidly undergoes
transition to the two-layered wake within a few diameters. Further downstream,
the two-layered wake becomes unstable and effectively sheds vortices forming a
secondary vortex street at a lower frequency. That wake state develops a visible
spatial oscillation, with a period approximately three times the second shedding
frequency. Thus, there are at least three frequencies that dominate the wake over
different spatial ranges. Note that the wake state shown in the image varies slowly
over time, because the primary and secondary shedding frequency are not multiples, as
is shown below. Of interest, the stability analysis of previous sections shows that the
near wake is stable to three-dimensional perturbations in this Reynolds number range.
An analysis using DMD is warranted because of the different frequencies involved
with the different spatial ranges of the wake. This was done using a sequence of 100
velocity fields separated by 0.5 time units to extract approximations to the Koopman
modes, i.e. the Ritz vectors (Rowley et al. 2009; Schmid 2010, 2011). The frequency
spectrum corresponding to the Ritz (eigen)values is shown in figure 19. The grey bar
at zero frequency (f0 ) corresponds to the mean mode. The black bars (f1 (St = 0.163))
correspond to the near-wake shedding and its harmonics, the downward striped bars
(f2 (St = 0.069)) correspond to the secondary shedding mode and its harmonics, and
the upward striped bar (f3 (St = 0.024)) is associated with the meandering of the
secondary wake towards the downstream end of the domain. In addition, the cross-
hashed bar is associated with the frequency difference f2 − f3 .
Reconstruction of the initial velocity field from the Ritz vectors is shown in
figure 20. The top image shows spanwise vorticity for the first velocity field of the
set. Figure 20(b) shows the zero-frequency ( f0 ) mode corresponding to the time
mean vorticity field. Below this, figure 20(c) shows the DMD reconstruction just
using the modes with frequency f0 + f1 + 2f1 + 3f1 + 4f1 + 5f1 , i.e. the mean field
together with the near-wake mode and its harmonics. This mode is localised in the
 Low-Reynolds-number wakes of elliptical cylinders 591

(a)

(b)

(c)

(d)

(e)

(f)

F IGURE 20. (Colour online) Reconstruction of wake for the Ar = 0.25 cylinder at Re = 150
from DMD modes. (a) Spanwise vorticity field for the original wake; (b) zero frequency
DMD mode ( f0 ); (c) reconstruction of the near wake (f0 + f1 + 2f1 + 3f1 + 4f1 + 5f1 );
(d) reconstruction of secondary wake ( f0 + f2 + 2f2 + 3f2 + 4f2 + 5f2 ); (e) low-order
reconstruction of the meandering secondary wake ( f1 + f2 + f3 + (f2 − f3 )); (f ) higher-order
reconstruction based on ( f1 + f2 + f3 + (f2 − f3 ) + 2f2 + 3f2 + 4f3 + 5f2 ). See the text for
further details.

near wake, as expected, and duplicates the actual vorticity field there extremely well.
The next image, figure 20(d), shows a reconstruction using modes corresponding to
frequencies f0 + f2 , which captures the secondary wake further downstream. This
shows (near-)zero oscillation amplitude in the near wake, where the BvK and
two-layered wakes reside. It persists further downstream, although weakened, to
enable reconstruction of the meandering secondary wake state. The penultimate
image corresponds to a reconstruction based on f0 + f2 + f3 + (f2 − f3 ), i.e. the mean
field together with the secondary wake mode, the lower-frequency meandering mode
and the difference between them. This reconstructs the meandering secondary wake
towards the end of the domain particular well, although near the secondary transition
the vortices are rather diffuse. The final image (figure 20f ) is a reconstruction
based on that combination together with the harmonics of the secondary wake mode
(f0 + f2 + f3 + (f2 − f3 ) + 2f2 + 3f2 + 4f2 + 5f2 ). This captures the wake downstream
from the development of secondary shedding. From the vorticity field shown in
figure 20(a), the secondary wake develops from merging of the upstream two-layer
vortices while those latter vortices are still distinct, i.e. before those vortices have
diffused to produce a locally time-independent wake. However, despite this, the ratio
of the frequencies of the primary to secondary wake is not an integer, as it would
be if the secondary wake vortices structures were obtained from direct merging
of the primary two-layer vortices. This ratio is St1 /St2 = 0.163/0.069 = 2.36. In
addition, the ratio of the secondary wake frequency to the meandering frequency is
St2 /St3 = 0.069/0.024 = 2.87, again not an integer, but still close to 3 so that the
meandering wavelength seems to be associated with groups of 3 vortices from the
secondary wake. The upshot of this is that the wake snapshot shown in figure 20(a) is
592 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

(a)

(b)

(c)

(d )

(e)

F IGURE 21. (Colour online) Development and reconstruction of wake for the Ar =
0.25 cylinder at Re = 150 at various times. (a) Spanwise vorticity field for the original
wake (t = 0); (b) vorticity field approximately one meandering period later (t = 43);
(c) DMD reconstruction at this time (t = 43) using all modes; (d) wake a further two
meandering periods later (t = 129); (e) DMD reconstruction at that time (t = 129). The
latter image shows that the near and far fields are reproduced reasonably accurately, but
the intermediate wake is less accurate. See the text for further details.

not representative of all times, but changes slowly with time. For instance, figure 21
shows the wake at a number of different times. The first two images ((a) and (b))
show the wake approximately one meandering period apart. Indeed, these two wakes
appear almost visually identical. The next image (c) shows that DMD reconstruction
using all of the modes for the same time, which verifies that the reconstruction
faithfully reproduces the actual velocity field. The next image (d) shows the wake
at another two meandering periods later from direct simulations, which is well past
the end time of the set of velocity fields used for the DMD analysis. This wake
looks considerably different for the initial part of the secondary wake. Reconstruction
of the wake through extrapolation up to this time using all of the DMD modes,
i.e. extrapolating to non-dimensional time t = 129 with DMD analysis based on the
time interval t 6 49.5, is shown in the last image (e). This captures the near wake
accurately and the far wake meandering reasonably well. It is less representative over
the intermediate range downstream of where the secondary shedding starts.
It is interesting to see how many modes are required to capture the near-wake
accurately. This is shown in figure 22. The top image shows the original vorticity
field. The next five images shows a reconstruction using the mean mode and mode
f1 together with zero to four harmonics of f1 . Clearly most of the detail is accurately
captured with just the first mode. Adding the next two harmonics improves the fidelity
substantially, especially over the very near wake. The addition of four harmonics
produces an image that is almost visually identical to the original, except for a very
small amount of noise close to the forming vortex.
Finally, DMD is used to examine the differences in the wake structure between two-
and three-dimensional simulations for the flat plate wake. The wake was examined
at Re = 140 and 180 corresponding to the three-dimensional simulations of figure 18.
In each case, 100 snapshots were used for the reconstruction spaced at 0.61 time
 Low-Reynolds-number wakes of elliptical cylinders 593

(a)

(b)

(c)

(d)

(e)

(f)

F IGURE 22. (Colour online) Reconstruction of the near wake using DMD modes. The
top image shows the wake at zero time. The following images shows the near wake
reconstruction based on the mean field together with mode f1 and zero to four harmonics:
(a) full field; (b) mode f1 ; (c) modes f1 + 2f1 ; (d) modes f1 + 2f1 + 3f1 ; (e) modes
f1 + 2f1 + 3f1 + 4f1 ; (f ) modes f1 + 2f1 + 3f1 + 4f1 + 5f1 .

units apart. The spanwise-averaged velocity was used for the analysis based on the
three-dimensional simulations. Figure 23 shows the change to the mean centreline
velocity determined from the mean DMD mode. At Re = 140, there is only a
minor change to the near-wake centreline velocity variation between the two- and
three-dimensional simulations. For the two-dimensional case, the velocity is close to
zero up to x/D ' 17 downstream, while the three-dimensional result shows a velocity
of considerably less than 10 % of the background velocity over the similar range.
Further downstream, unsurprisingly, the two-dimensional velocity recovers towards
the free-stream value considerably faster. The higher-Reynolds-number results are
considerably different. The flow reversal is not observed in the three-dimensional
case, and indeed the velocity recovery in the three-dimensional case is rapid. These
observations are in accord with the DMD reconstructions shown at the right. They
show the near-wake structure using the mean mode together with the DMD mode with
the near-wake frequency and its first four harmonics. At the lower Reynolds number
of Re = 140, even though transition occurs at Re = 116, the two- and three-dimensional
solution-based reconstructions are very similar, indicating that the mean distortion of
594 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan

(a) 1.00
Mean centreline velocity (b)

0.75
2D
0.50 3D
0.25

−0.25

−0.50
0 10 20 30 40

1.00
Mean centreline velocity

0.75

0.50
3D
2D
0.25

−0.25

−0.50
0 5 10 15 20 25 30

F IGURE 23. (Colour online) (a) Variation of the centreline velocity in the wake of a
flat plate for Re = 140 (top) and Re = 180 (bottom). (b) Reconstruction of the near-wake
vorticity field from DMD modes. The mean mode and the mode corresponding to the
near-wake frequency, together with its first four harmonics, are used for the reconstruction.
The top image of each set uses the two-dimensional fields and the lower one uses the
three-dimensional simulation fields. Again, the upper and lower image sets correspond to
Re = 140 and Re = 180, respectively.

the two-dimensional vortex rollers is little affected by the three-dimensionality of the

wake. This is not true at the higher Reynolds number of Re = 180, where is it clear
that the three-dimensional reconstruction shows diffuse near-wake vortices relative
to the compact two-dimensional wake vortices. This is consistent with the stronger
induced roller distortion in the near wake at the higher Reynolds number.

3.11. Fourier analysis of centreline velocity

The dominant wake frequencies were further analysed using standard Fourier analysis
of the cross-stream velocity along the wake centreline. For the same case as examined
above for the DMD analysis, i.e. Ar = 0.25, Re = 150, figure 24(a) shows the spectral
power of the cross-stream velocity component at various downstream distances. This
is summarised in figure 24(b) which shows the dominant frequency components as a
function of downstream distance. The velocity was sampled at 1t = 0.01 time units
to produce N = 217 = 131 072 measurements. This set was then processed using a fast
Fourier transform (FFT) to produce these plots. This gives a frequency resolution of
1St = 2/(N1t) = 0.0015, with the data set consisting of more than 200 near-wake
 Low-Reynolds-number wakes of elliptical cylinders 595

(a) (b) 0

−1
log10(Power)

−2

10 −3
20
−4
100
200 −5
0 0.2 0.4 0.6 0.8 1.0 0 50 100 150 200 250 300
St Distance downstream

F IGURE 24. Spectral analysis of wake for the Ar = 0.25 cylinder at Re = 150.
(a) Power spectra at different downstream distances. (b) Power of the dominant frequency
components as a function of downstream distance.

(a) 0.18 (b) 40

Stnear wake
0.16
30
0.14

St 0.12 20

0.10 Stsecondary wake

10
0.08

0.06 0
130 140 150 160 170 180 190 130 140 150 160 170 180 190
Re Re

F IGURE 25. (a) Near-wake and secondary-wake frequencies as a function of Reynolds

number for the Ar = 0.25 cylinder. (b) Downstream position at which the wake undergoes
transition from the near wake to the secondary wake.

shedding cycles. The principal peaks corresponding to the near-wake, secondary wake
and meandering frequencies (St1 = 0.1630, St2 = 0.0689, St3 = 0.0251) match those
found from the 100 sample DMD analysis (0.1626, 0.0689, 0.0243, respectively) to
within the frequency resolution of the Fourier analysis. The transition from the near
wake to the secondary wake is further quantified in figure 25. The left-hand figure
shows the variation of the near- and secondary-wake frequencies as a function of
Reynolds number. For lower Reynolds number, the secondary frequency is less than
half of the near-wake frequency. The ratio is close to 2 for Re ' 170. Thus, again,
it is clear that the secondary wake does not result from direct merging of a discrete
number of near-wake vortices. At higher Reynolds numbers, for example Re = 160,
the secondary wake vortices merge further downstream resulting in a lowering of the
dominant wake frequency from St = 0.0733 to St = 0.0565. This occurs a long way
downstream at x/D ' 250. The transition from the near wake to the secondary wake
occurs progressively closer to the cylinder. It is quantified in figure 25 as the position
where the spectral power corresponding to the near- and secondary-wake frequencies
are equal.
596 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan
4. Conclusions
The critical Reynolds numbers and Strouhal frequencies for the initial transition
to unsteady periodic flow have been accurately determined for elliptical cylinders of
different aspect ratios. These results correlate well with previous studies when the
effects of blockage of previous studies are taken into account. The two-dimensional
simulations spanned Reynolds numbers up to 200, corresponding to the upper limit
of two-dimensional shedding for the circular cylinder. This produces a rich variety
of wake states. As the aspect ratio is lowered and the Reynolds number is increased,
the wake undergoes a series of changes that are not normally seen with the circular
cylinder wake. In the most extreme cases, the near-wake BvK shedding rapidly shifts
to a two-layered structure within a few diameters downstream. Further downstream,
the two-layered wake can become unstable, with an oscillation developing and a
secondary aligned vortex street the result. This may form either downstream of
where the two-layered vortices diffuse and merge due to a hydrodynamic instability
(Durgin & Karlsson 1971) to reach a time-independent state, or prior to this,
when the two-layered vortices are still discrete. In either case, the near-wake and
secondary-wake frequencies are not commensurate so that the secondary-wake
vortices are not the result of merging of a discrete number of two-layer vortices.
The secondary wake initially takes the form of a typical BvK vortex street, but as
the Reynolds number is increased, that street develops (near-)periodic oscillations and
even chaotic behaviour with further merging of vortices, the formation of identifiable
vortex pairs and substantial increases in the wake width. It is clear that decreasing the
aspect ratio results in a substantial increase in the circulation feeding into the wake,
due to the increased speed up of the flow as it passes around the top of the cylinder.
For a flat plate, this circulation is approximately a factor of two greater than for a
circular cylinder at Re = 200, hence the wake vortices are stronger and also larger.
The centreline wake deficit variation with Reynolds number has also been quantified.
The downstream variation is surprisingly different from that of a circular cylinder
once the two-layer wake develops. This leads to a very low mean velocity due to
the combined influence of the offset negative and positive lines of vortices. Even
for the Ar = 0.5 cylinder, the mean centreline velocity is close to zero downstream
to approximately 20 diameters. For the flat plate at Re = 150 and above, there is a
significant reverse flow further downstream from the immediate near wake (as was
observed by Saha (2013)).
The Strouhal/Reynolds number variations from the two-dimensional simulations
are quantified. These show a considerably more complex behaviour than that for a
circular cylinder. The agreement with experimental results is favourable, when the
flow is not three-dimensional. The onset of three-dimensional flow is also predicted.
In this case, decreasing the aspect ratio decreases the Reynolds number at which
the BvK near wake undergoes downstream transition to the two-layered wake.
This strongly affects the transition scenario. For aspect ratios of 0.5 and above,
the initial transition is equivalent to the mode A transition for a circular cylinder,
even though the critical Reynolds number decreases substantially as the aspect
ratio decreases. At Ar = 0.25, the initial transition is also similar to mode A. The
critical Reynolds number is approximately 88. However, increasing the Reynolds
number leads to such a change in the two-dimensional wake structure that mode A
is no longer unstable beyond Re ' 125. Thus, at least the near-wake returns to a
two-dimensional state. This two-dimensional state remains stable until Re = 185,
when another three-dimensional mode with the same spatiotemporal symmetry as
mode A but a much shorter wavelength (λ = 1.3D) becomes unstable. However,
 Low-Reynolds-number wakes of elliptical cylinders 597
the wake shows little resemblance to a classical BvK vortex street at this Reynolds
number. For the flat plate, the transition to the two-layered wake state occurs at an
even lower Reynolds number. This seems to prevent mode A becoming unstable at all.
Instead, at Re = 116, a quasi-periodic three-dimensional mode becomes unstable, and
at Re = 141 another mode which has the same period as the base flow also becomes
unstable. This second mode appears to be equivalent to the mode causing the second
three-dimensional transition for the Ar = 0.25 cylinder. Three-dimensional simulations
at Re = 140 and Re = 180 confirm the three-dimensional nature of the wake at these
Reynolds numbers and show that the major distortion of the two-dimensional vortex
rollers occurs towards the end of the two-layered wake region. The three-dimensional
wake has a higher Strouhal number than its two-dimensional counterpart.
DMD was also applied to the Ar = 0.25 wake at Re = 150. This case corresponds to
when the wake shows quite distinct temporal behaviours over different spatial ranges.
The analysis using only a short time sequence of 100 snapshots was able to extract
the spatially localised wake components together with accurate determination of their
corresponding frequencies. Those frequencies agree with the dominant frequencies
obtained from Fourier analysis of the cross-stream velocity on the centreline using
much longer and finer temporal data sets. The DMD analysis allows an accurate
reconstruction of the wake state at any time using just the dominant modes and their
harmonics.

Acknowledgements
The authors acknowledge computing-time support from the Victorian Life Sciences
Computation Initiative (VLSCI) and the National Computational Infrastructure
(NCI). The authors acknowledge financial support from Australian Research Council
Discovery Project grants DP110102141 and DP130100822.

REFERENCES

A LBAREDE , P. & M ONKEWITZ , P. A. 1992 A model for the formation of oblique shedding and
Chevron patterns in cylinder wakes. Phys. Fluids A 4 (4), 744–756.
A LEKSYUK , A. I., S HKADOVA , V. P. & S HKADOV, V. Y. 2012 Formation, evolution, and decay of
a vortex street in the wake of a streamlined body. Moscow Univ. Mech. Bull. 67 (3), 53–61.
BARKLEY, D. & H ENDERSON , R. D. 1996 Three-dimensional Floquet stability analysis of the wake
of a circular cylinder. J. Fluid Mech. 322, 215–241.
B EHARA , S. & M ITTAL , S. 2010 Wake transition in flow past a circular cylinder. Phys. Fluids 22
(11), 114104.
B ÉNARD , H. 1908 Formation de centres de giration à l’àrriere d’un obstacle en mouvement. C. R.
Acad. Sci. Paris 147, 839–842.
C HAURASIA , H. K. & T HOMPSON , M. C. 2012 Three-dimensional instabilities in the boundary-layer
flow over a long rectangular plate. J. Fluid Mech. 681, 411–433.
C HORIN , A. J. 1968 Numerical solution of the Navier–Stokes equations. Maths Comput. 22, 745–762.
C IMBALA , J. M. 1984 Large structure in the far wakes of two-dimensional bluff bodies. PhD thesis,
Graduate Research Laboratories.
C IMBALA , J. M., N AGIB , H. M. & ROSHKO , A. 1988 Large structure in the far wakes of two-
dimensional bluff bodies. J. Fluid Mech. 190, 265–298.
D URGIN , W. W. & K ARLSSON , S. K. F. 1971 On the phenomenon of vortex street breakdown.
J. Fluid Mech. 48 (03), 507–527.
D USEK , J., L E G AL , P. & F RAUNIE , D. P. 1994 A numerical and theoretical study of the first Hopf
bifurcation in a cylinder wake. J. Fluid Mech. 264, 59–80.
598 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan
E ISENLOHR , H. & E CKELMANN , H. 1989 Vortex splitting and its consequences in the vortex street
wake of cylinders at low Reynolds number. Phys. Fluids A 1 (2), 189–192.
FAGE , A. & J OHANSEN , F. C. 1927 On the flow of air behind an inclined flat plate of infinite span.
Proc. R. Soc. Lond. A 116 (773), 170–197.
G RIFFITH , M. D., L EWEKE , T., T HOMPSON , M. C. & H OURIGAN , K. 2008 Steady inlet flow in
stenotic geometries: convective and absolute instabilities. J. Fluid Mech. 616, 111–133.
G RIFFITH , M. D., L EWEKE , T., T HOMPSON , M. C. & H OURIGAN , K. 2010 Convective instability
in steady stenotic flow: optimal transient growth and experimental observation. J. Fluid Mech.
655, 504–514.
H ENDERSON , R. D. 1995 Details of the drag curve near the onset of vortex shedding. Phys. Fluids
7 (9), 2102–2104.
H ENDERSON , R. D. 1997 Nonlinear dynamics and pattern formation in turbulent wake transition.
J. Fluid Mech. 352 (1), 65–112.
JACKSON , C. P. 1987 A finite-element study of the onset of vortex shedding in flow past variously
shaped bodies. J. Fluid Mech. 182, 23–45.
J EONG , J. & H USSAIN , F. 1995 On the identification of a vortex. J. Fluid Mech. 285, 69–94.
J OHNSON , S. A., T HOMPSON , M. C. & H OURIGAN , K. 2004 Predicted low frequency structures in
the wake of elliptical cylinders. Eur. J. Mech. (B/Fluids) 23 (1), 229–239.
K ARASUDANI , T. & F UNAKOSHI , M. 1994 Evolution of a vortex street in the far wake of a cylinder.
Fluid Dyn. Res. 14 (6), 331–352.
K ÁRMÁN , T H . V. 1911 Über den Mechanismus des Widerstandes, den ein bewegter Körper in einer
Flüsseigkeit erfährt. Gött. Nachr. 509–511.
K ARNIADAKIS , G. E., I SRAELI , M. & O RSZAG , S. A. 1991 High-order splitting methods for the
incompressible Navier–Stokes equations. J. Comput. Phys. 97, 414–443.
K ARNIADAKIS , G. E. & S HERWIN , S. J. 2005 Spectral/HP Methods for Computational Fluid
Dynamics. Oxford University Press.
K ARNIADAKIS , G. E. & T RIANTAFYLLOU , G. S. 1992 Three-dimensional dynamics and transition
to turbulence in the wake of bluff objects. J. Fluid Mech. 238, 1–30.
K UMAR , B. & M ITTAL , S. 2006 Effect of blockage on critical parameters for flow past a circular
cylinder. Intl J. Numer. Meth. Fluids 50 (8), 987–1001.
K UMAR , B. & M ITTAL , S. 2012 On the origin of the secondary vortex street. J. Fluid Mech. 711,
641–666.
L E G AL , P., NADIM , A. & T HOMPSON , M. C. 2001 Hysteresis in the forced Stuart–Landau equation:
application to vortex shedding from an oscillating cylinder. J. Fluids Struct. 15, 445–457.
L EONTINI , J. S., L O JACONO , D. & T HOMPSON , M. C. 2013 Wake states and frequency selection
of a streamwise oscillating cylinder. J. Fluid Mech. 730, 162–192.
L EONTINI , J. S., T HOMPSON , M. C. & H OURIGAN , K. 2007 Three-dimensional transition in the
wake of a transversely oscillating cylinder. J. Fluid Mech. 577, 79–104.
M AMUN , C. K. & T UCKERMAN , L. S. 1995 Asymmetry and Hopf-bifurcation in spherical Couette
flow. Phys. Fluids 7, 80–91.
M ILLER , G. D. & W ILLIAMSON , C. H. K. 1994 Control of three-dimensional phase dynamics in a
cylinder wake. Exp. Fluids 18 (1), 26–35.
M ODI , V. J. & D IKSHIT, A. K. 1975 Near-wakes of elliptic cylinders in subcritical flow. AIAA J.
13 (4), 490–497.
M ODI , V. J. & W ILAND , E. 1970 Unsteady aerodynamics of stationary elliptic cylinders in subcritical
flow. AIAA J. 8 (10), 1814–1821.
M ONKEWITZ , P. A. 1988 The absolute and convective nature of instability in two-dimensional wakes
at low Reynolds numbers. Phys. Fluids 31 (5), 999–1006.
NAJJAR , F. M. & BALACHANDAR , S. 1998 Low-frequency unsteadiness in the wake of a normal
flat plate. J. Fluid Mech. 370, 101–147.
N ORBERG , C. 1987 Effects of Reynolds number and a low-intensity free stream turbulence on the
flow around a circular cylinder. PhD thesis, Chalmers University of Technology.
OTA , T., N ISHIYAMA , H. & TAOKA , Y. 1987 Flow around an elliptic cylinder in the critical Reynolds
number regime. Trans. ASME J. Fluids Engng 109 (2), 149–155.
 Low-Reynolds-number wakes of elliptical cylinders 599
R ADI , A., T HOMPSON , M. C., S HERIDAN , J. & H OURIGAN , K. 2013 From the circular cylinder
to the flat plate wake: the variation of Strouhal number with Reynolds number for elliptical
cylinders. Phys. Fluids 25, 101706.
R AO , A., L EONTINI , J. S., T HOMPSON , M. C. & H OURIGAN , K. 2013a Three-dimensionality in
the wake of a rotating cylinder in a uniform flow. J. Fluid Mech. 717, 1–29.
R AO , A., T HOMPSON , M. C., L EWEKE , T. & H OURIGAN , K. 2013b Dynamics and stability of the
wake behind tandem cylinders sliding along a wall. J. Fluid Mech. 722, 291–316.
ROSHKO , A. 1954 On the Drag and Shedding Frequency of Two-Dimensional Bluff Bodies. National
Aeronautics and Space Administration.
ROWLEY, C. W., M EZI Ć , I., BAGHERI , S., S CHLATTER , P. & H ENNINGSON , D. S. 2009 Spectral
analysis of nonlinear flows. J. Fluid Mech. 641, 115–127.
RYAN , K., T HOMPSON , M. C. & H OURIGAN , K. 2005 Three-dimensional transition in the wake of
elongated bluff bodies. J. Fluid Mech. 538, 1–29.
S AHA , A. K. 2007 Far-wake characteristics of two-dimensional flow past a normal flat plate. Phys.
Fluids 19, 128110.
S AHA , A. K. 2013 Direct numerical simulation of two-dimensional flow past a normal flat plate.
J. Engng Mech. ASCE 12, 1894–1901.
S CHMID , P. J. 2010 Dynamic mode decomposition of numerical and experimental data. J. Fluid
Mech. 656, 5–28.
S CHMID , P. J. 2011 Application of the dynamic mode decomposition to experimental data. Exp.
Fluids 50, 1123–1130.
S HINTANI , K., U MEMURA , A. & TAKANO , A. 1983 Low-Reynolds-number flow past an elliptic
cylinder. J. Fluid Mech. 136 (1), 277–289.
S TEWART, B. E., T HOMPSON , M. C., L EWEKE , T. & H OURIGAN , K. 2010 The wake behind a
cylinder rolling on a wall at varying rotation rates. J. Fluid Mech. 648, 225–256.
S TROUHAL , V. 1878 Üeber eine besondere Art der Tonerregung. Ann. Phys. 241 (10), 216–251.
S TRYKOWSKI , P. J. & S REENIVASAN , K. R. 1990 On the formation and suppression of vortex
‘shedding’ at low Reynolds numbers. J. Fluid Mech. 218, 71–107.
TANEDA , S. 1959 Downstream development of the wakes behind cylinders. J. Phys. Soc. Japan 14
(6), 843–848.
T HOMPSON , M. C., H OURIGAN , K., C HEUNG , A. & L EWEKE , T. 2006a Hydrodynamics of a particle
impact on a wall. Appl. Math. Model. 30, 1356–1369.
T HOMPSON , M. C., H OURIGAN , K., RYAN , K. & S HEARD , G. J. 2006b Wake transition of two-
dimensional cylinders and axisymmetric bluff bodies. J. Fluids Struct. 22 (6–7), 793–806.
T HOMPSON , M. C., H OURIGAN , K. & S HERIDAN , J. 1996 Three-dimensional instabilities in the
wake of a circular cylinder. Exp. Therm. Fluid Sci. 12 (2), 190–196.
T HOMPSON , M. C., L EWEKE , T. & H OURIGAN , K. 2007 Sphere-wall collisions: vortex dynamics
and stability. J. Fluid Mech. 575, 121–148.
T HOMPSON , M. C., L EWEKE , T. & W ILLIAMSON , C. H. K. 2001 The physical mechanism of
transition in bluff body wakes. J. Fluids Struct. 15 (3–4), 607–616.
T SUBOI , K. & O SHIMA , Y. 1985 Merging of two-dimensional vortices by the discrete vortex method.
J. Phys. Soc. Japan 54 (6), 2137–2145.
VOROBIEFF , P., G EORGIEV, D. & I NGBER , M. S. 2002 Onset of the second wake: dependence on
the Reynolds number. Phys. Fluids 14 (7), L53–L56.
W ILLIAMSON , C. H. K. 1988 The existence of two stages in the transition to three-dimensionality
of a cylinder wake. Phys. Fluids 31 (11), 3165–3168.
W ILLIAMSON , C. H. K. 1989 Oblique and parallel modes of vortex shedding in the wake of a
circular cylinder at low Reynolds numbers. J. Fluid Mech. 206, 579–627.
W ILLIAMSON , C. H. K. 1996a Three-dimensional wake transition. J. Fluid Mech. 328, 345–407.
W ILLIAMSON , C. H. K. 1996b Vortex dynamics in the cylinder wake. Annu. Rev. Fluid Mech. 28
(1), 477–539.
W ILLIAMSON , C. H. K. & P RASAD , A. 1993 A new mechanism for oblique wave resonance in the
far wake. J. Fluid Mech. 256, 269–313.
600 M. C. Thompson, A. Radi, A. Rao, J. Sheridan and K. Hourigan
YANG , D., N ARASIMHAMURTHY, V. D., P ETTERSEN , B. & A NDERSSON , H. I. 2012 Three-
dimensional wake transition behind an inclined flat plate. Phys. Fluids 24 (9), 094107.
Z HANG , H.-Q., F EY, U., N OACK , B. R., K ONIG , M. & E CKELMANN , H. 1995 On the transition of
the cylinder wake. Phys. Fluids 7 (4), 779–794.